CVE-2024-0095

9.0 CRITICAL

📋 TL;DR

CVE-2024-0095 is a log injection vulnerability in NVIDIA Triton Inference Server that allows attackers to inject forged logs and executable commands by manipulating log entries. This could lead to remote code execution, privilege escalation, or data tampering. Organizations using vulnerable versions of NVIDIA Triton Inference Server on Linux or Windows are affected.

💻 Affected Systems

Products:
  • NVIDIA Triton Inference Server
Versions: All versions prior to 24.01
Operating Systems: Linux, Windows
Default Config Vulnerable: ⚠️ Yes
Notes: Affects both Linux and Windows deployments of Triton Inference Server. The vulnerability exists in the logging functionality.

📦 What is this software?

⚠️ Risk & Real-World Impact

🔴

Worst Case

Full system compromise with remote code execution leading to complete control of the Triton server, data exfiltration, and lateral movement within the network.

🟠

Likely Case

Denial of service through log manipulation or limited code execution in the Triton server context, potentially leading to inference service disruption.

🟢

If Mitigated

Log injection attempts are detected and blocked, with no successful exploitation due to proper input validation and logging controls.

🌐 Internet-Facing: HIGH
🏢 Internal Only: MEDIUM

🎯 Exploit Status

Public PoC: ✅ No
Weaponized: UNKNOWN
Unauthenticated Exploit: ⚠️ Yes
Complexity: LOW

The vulnerability allows injection of arbitrary data as log entries, which could be leveraged for command execution. No public exploit code has been identified at this time.

🛠️ Fix & Mitigation

✅ Official Fix

Patch Version: 24.01 or later

Vendor Advisory: https://nvidia.custhelp.com/app/answers/detail/a_id/5546

Restart Required: Yes

Instructions:

1. Download NVIDIA Triton Inference Server version 24.01 or later from NVIDIA's official channels. 2. Stop the Triton Inference Server service. 3. Install the updated version following NVIDIA's installation guide. 4. Restart the Triton Inference Server service.

🔧 Temporary Workarounds

Restrict Log Input Sources

all

Implement input validation and sanitization for all log entry sources to prevent injection of arbitrary data.

# Configure Triton server to validate log inputs
# Review and modify logging configuration to restrict input sources

Network Segmentation

linux

Isolate Triton Inference Server from untrusted networks and restrict access to authorized clients only.

# Configure firewall rules to restrict access
iptables -A INPUT -p tcp --dport 8000 -s trusted_ip_range -j ACCEPT
iptables -A INPUT -p tcp --dport 8000 -j DROP

🧯 If You Can't Patch

  • Implement strict network access controls to limit Triton server exposure to only trusted internal networks.
  • Deploy web application firewall (WAF) rules to detect and block log injection patterns in requests to the Triton server.

🔍 How to Verify

Check if Vulnerable:

Check the Triton Inference Server version. If it's earlier than 24.01, the system is vulnerable.

Check Version:

tritonserver --version

Verify Fix Applied:

Verify the Triton Inference Server version is 24.01 or later and test logging functionality with controlled input to ensure no injection occurs.

📡 Detection & Monitoring

Log Indicators:

  • Unusual log entries containing executable code or command syntax
  • Log entries with unexpected characters or patterns suggesting injection attempts

Network Indicators:

  • Unusual HTTP requests to Triton server logging endpoints
  • Requests containing log injection patterns in payloads

SIEM Query:

source="triton_server" AND (message="*exec*" OR message="*cmd*" OR message="*inject*")

🔗 References

📤 Share & Export